Page 15 - TECH MAGAZINE CSE

AN Technical Magazine
VESHAN Maharaja Agrasen Institute of Technology

The steps below can be used to characterise a typical GNN layer:

1.Message passing is the process by which each node gets messages from its neighbours.
These messages are predicated on the present features of the neighbour nodes and
potentially the edge features.
2.Aggregation: The node combines the messages it gets from its neighbours, usually with
the help of functions like max, mean, or sum.
3.Update: The node uses the combined messages to update its own feature representation,
usually via a fully connected layer of a neural network.

Applications of GNNs
a. Social Networks:
• Node Classification: Group individuals into categories by predicting their attributes based
on their relationships.

• Link Prediction: Estimate the probability of creating new connections (such as friend
recommendations on social media).
b. Chemistry and Molecular Biology:
• Mole Classification: By treating atoms as nodes and bonds as edges, GNNs are utilised to
predict molecular characteristics (such as toxicity and solubility).
• Drug Discovery: GNNs assist in simulating the interactions between chemicals and
proteins in order to identify possible therapeutic candidates.
c. Recommendation Systems: GNNs are used to enhance recommendation systems by
using graph structure to provide better recommendations by modelling users and objects
as nodes and interactions as edges.
d. Knowledge Graphs:
• Entity and Relation Prediction: GNNs are able to detect relationships between entities in
a knowledge base, for example, and predict missing facts in knowledge graphs.
e. Transportation and Traffic Networks: GNNs are used to model traffic patterns in road
networks, where roads are represented by edges and intersections by nodes, enabling
more accurate predictions of traffic flow.

Conclusion

GNNs are an essential component of contemporary AI systems due to their capacity to
recognise and interpret the intricate connectivity seen in these applications.
In domains where the connections between data points are just as significant as the
individual points themselves, GNNs effectively train and predict by extending the
capabilities of neural networks to graph-structured data. GNNs aid in predicting groups
and linkages in social networks. They improve user-item interaction modelling in
recommendation systems to produce tailored recommendations and offer insights into
molecular and genetic relationships in biological networks.